This assignment is for ETC5521 Assignment 1 by Team goanna comprising of NULL, NULL, and Dea Avega Editya.

1 Introduction and motivation

Bushfires that raged in Australia from September 2019 to early 2020 captured the attention of people worldwide, especially the out-of-control bushfires in Victoria and New South Wales(NSW). However, looking at the Australian bushfires’ history data, we can find that Australia suffers from different degrees and quantities of bushfires every year. It is exciting to explore the link between Australian fire history and climate change. What climatic conditions caused the frequent occurrence of the bushfires in Australia?
In this analysis, R is the main tool for data cleaning and analysis.

The rest of the analysis proceeds as follows. Section 2 presents the data description. Section 4 details the findings in data analysis. The limitations of the analysis are presented in Section 3. Finally, Section 5 provides the conclusions of this analysis.

1.1 Research questions

This analysis aims to explore three secondary questions:
- When and where were the most widespread fires burning?
- Was temperature to be blamed for bushfires in Australia?
- Was rainfall to be blamed for bushfires in Australia?

2 Data description

This section mainly introduces the data, data sources and data description.
There are three data sets used on this analysis, and the cleaned data is obtained from GitHub tidytuesday.

2.1 Australia fire data

The fire data source is from NASA(NASA 2020) “Active Fires Dataset” via the MODIS fire product collection. MODIS active fire product is detected the fire data in every 5 minutes and collected through each tile with horizontal and vertical coordinate.The data contains the fire information of Australia with 5101817 observations from 2000-11-01 to 2020-01-05.

Table 2.1: Australia fire data
Variable Description
latitude Center of 1km fire pixel but not necessarily the actual location of the fire as one or more fires can be detected within the 1km pixel.
longitude Center of 1km fire pixel but not necessarily the actual location of the fire as one or more fires can be detected within the 1km pixel.
brightness Channel 21/22 brightness temperature of the fire pixel measured in Kelvin.
scan The algorithm produces 1km fire pixels but MODIS pixels get bigger toward the edge of scan. Scan and track reflect actual pixel size.
track The algorithm produces 1km fire pixels but MODIS pixels get bigger toward the edge of scan. Scan and track reflect actual pixel size.
acq_date Date of MODIS acquisition.
act_time Time of acquisition/overpass of the satellite (in UTC).
satellite A = Aqua and T = Terra.
confidence This value is based on a collection of intermediate algorithm quantities used in the detection process. It is intended to help users gauge the quality of individual hotspot/fire pixels. Confidence estimates range between 0 and 100% and are assigned one of the three fire classes (low-confidence fire, nominal-confidence fire, or high-confidence fire).
version Version identifies the collection (e.g. MODIS Collection 6) and source of data processing: Near Real-Time (NRT suffix added to collection) or Standard Processing (collection only). ‘6.0NRT’ - Collection 6 NRT processing.’6.0’ - Collection 6 Standard processing. Find out more on collections and on the differences between FIRMS data sourced from LANCE FIRMS and University of Maryland.
dbright_t31 Channel 31 brightness temperature of the fire pixel measured in Kelvin.
frq Depicts the pixel-integrated fire radiative power in MW (megawatts).
day_night D = Daytime, N = Nighttime

2.2 Climate data

For climate data, temperature and rainfall was gathered from the Australian Bureau of Meterology (BoM), of which is used the weather station to measure a variety of aspects of the weather.

In order to maintain the complete information in six cities, we added some missing values for Brisbane and Canberra from the source website and cleaned them to obtain the data from 1968-01-01 to 2017-12-31 in Canberra and from 1893-01-01 to 1998-12-31 in Brisbane. The new rainfall data is constructed by the old one and supplemental one and then become the final rainfall data set.

The temperature data is divided into two parts, with the cleaned data obtained directly from GitHub tidytuesday, which is from 1910-01-01 to 2019-05-31. And we found the time range of the cleaned temperature data cannot match the fire data, in order to further analysis of the relationship between the fire and the temperature, we decided to download the new temperature data from the source website, and cleaned it to obtain the new data from 2019-06-01 to 2020-01-05.

Also, there are seven weather stations were chosen, based on seven Australian cities such as Perth, Adelaide, Melbourne, Sydney, Brisbane, Port Lincoln and Canberra.

The climate data structure is shown as the below tables. In this analysis, date, temperature and temp_type variables for temperature data set were mainly used, and year, city_name and rainfall variables for rainfall data set as well.

Table 2.2: Temperature data
Variable Class Description
city_name character City Name
date double Date
temperature double Temperature in Celsius
temp_type character Temperature type (min/max daily)
site_name character Actual site/weather station
Table 2.3: Rainfall data
Variable Class Description
station_code character Station Code
city_name character City Name
year double year
month character month
day character day
rainfall double Trainfall in millimeters
period double how many days was it collected across
quality character Certified quality or not
lat double latitude
long double longitude
station_name character Station Name

3 Limitations of analysis

This section mainly introduces two main limitations of this analysis:
- There is no regional division in the fire data. We have indistinctly divided seven states or regions according to Australia’s longitude and latitude, which may lead to bias in the analysis results for location.
- The temperature and rainfall data include only some major cities, and comparing with the fire data, the sample is too small. Therefore, the analysis of the link between fires and climate data in Section 4 is not accurate and will cause deviation in results.

4 Analysis and findings

4.1 Climate Conditions

4.1.1 Where does rainfall occur the most in Australia?

4.1.2 Where is the hottest climate in Australia?

4.2 Bushfires

The Australian climate is generally hot and dry, bushfires can occur at anytime of the year for the most of regions(Australia 2020). However, are there more bushfires prefer happened in the summer or winter?

In this section, we will discuss it by analyzing the bushfire data for the past 20 years.

4.2.1 In which months are bushfires burning?

Figure 4.1: Yearly Australian fires from 2001 until 2020

From Figure 4.1, there are some unusual phenomena. Firstly, we can see that the bushfires trend line in 2019 is quite different from other years.

When the trend in other years (i.e. 2011 and 2012) seems to reach a peak in August and October respectively, the trend of bushfire in 2019 is still growing until December 2019. Unfortunately, we do not have additional data to see the continuity of the 2019 trend line in 2020.

According to the figure, the most severe bushfire occurs in 2011 with the number of spots reach 120.348 fire spots. We will look further into 2011 in the figure below.

–>

Figure 4.2: Average monthly rainfall from 2001

As can be seen from figure 4.2, in 2011 (which is the worst year)

Also, it’s interesting to find that bushfires are more likely to occur in May, and month from August to November, especially October. However, August to October is the winter to spring period in Australia. It seems that the frequent occurrence of bushfires during this period mainly due to dryness, which makes forest fuels more likely to catch fire(Sullivan et al. 2012). Moreover, comparing with Figure 4.2, it can be easy to find that the driest months are September, October and May, which coincided with the peak of the bushfires.

4.2.2 Where are the bushfires burning?

In this part, we will focus on the relationship between bushfires and geographic locations. Firstly, we divided bushfire data into different states or regions according to the latitude and longitude, because the data downloaded from NASA does not contain information for states or regions. Since the borders of NSW and Victoria are not easily distinguished, so we finally putting them together as one region.

 The number of bushfire in different states or regions in the past 20 years

Figure 4.3: The number of bushfire in different states or regions in the past 20 years

Figure 4.3: The number of bushfire in different states or regions in the past 20 years

Furthermore, we also draw a line that shows national average number of bushfires during 20 years period. Hence, we can see which states/territories that have more bushfires compared to national average. In addition, we are interested to see particular year of 2011 (the worst year) and 2019 (the different pattern).

Figure 4.3 shows us that, in general, bushfires are mostly occured in the Northern Territory(NT) and West Australia (WA), possibly because both areas are more likely prone to drought (we will check this assumption in the following section) and they also have large forests area (according to Australia’s Department of Agriculture)[https://www.agriculture.gov.au/abares/forestsaustralia/australias-forests#:~:text=%E2%80%8BQueensland%20has%20the%20largest,up%20much%20of%20the%20balance].

In 2011, NT has very high number of bushfires compared to other states. However, latest massive bushfires in 2019 is mostly occured in area comprises of Victoria and New South Wales. Figure 4.4 shows fire spots from the latest bushfires (December 2019 to early January 2020), and we can find that it is mainly concentrated in Victoria and NSW.

Based on the analysis, we will look at the causes of bushfires in Australia. Are there association between climate conditions to the bushfire? We will explore the story of climate conditions (temperature and rainfall level) in section 4.3.1 and 4.3.2.

Figure 4.4: The Fire Spots during 2019-12-29 until 2020-01-05(the darker the color, the more serious the fire)

4.3 How does temperature and rainfall affect the number of bushfires?

4.3.1 Positive association with temperature and bushfires

This section will be focus on finding clues to answer research question of whether higher temperature contributes to higher frequency of bushfires occurences. Our base assumption is that these two variables have a positive association which means that if the temperature increase the number of fire spots should also increase and vice versa.

Indian Ocean Dipole: positive phase. Source from Australian BOM.

Figure 4.5: Indian Ocean Dipole: positive phase. Source from Australian BOM.

In order to see the association between temperature and bushfire, we will produce a plot that compares these two variables side by side. To make a more visible pattern, temperature variable will be plotted as a temperature difference from the average temperature during period of observations (2000 - 2020). Mathematically, the temperature difference is formulated as below:

\[ Average\;Temperature\;Difference_n = Annual\;Temperature_n\; - Average\;in\;Period\;of\;Observations\]

Figure 4.6: The plot for the difference between the average temperature of 1961-1990(as baseline) and the annual average temperature for each year from 1910 to 2019, calculated by daily maximum temperature

On the other hand, since the average number of bushfires (fire spots) has different scale with the temperature, we will scale down the annual average bushfires using log 10 and square root. By doing so, we can put the two variables in a single plot to see the patterns of each variables.

As we can see from figure 4.6, the annual temperature trend seems to rise since 2000 until 2019. Likewise, the annual bushfires also slighlty increase during observed years. Therefore, we can see a positive association between these two variables, although, the association seems not really strong.

As an example, we can see that in 2011 the average bushfires rise sharply (marked by black dashed line) while the temperature only a bit increase from 2010. Another example of the weak association is also shown in the trends during 2015 until 2020. In that period, average bushfires seems to slighlty increase whereas the temperature actually rises significantly. Further analysis on correlation between temperature and bushfires will be discussed in the last section of this paper.

Figure 4.7: Annual total fires trends from 2001 to 2019

4.3.2 Negative association with rain and bushfires

Has the less rainfall affected the more bushfires in Australia? Or has the more rainfall influenced the less bushfires? Because of the location of Australia, the rainfall in Australia is highly variable, which is strongly influenced by global climate system phenomena such as El Niño, La Niña, and IOD. Despite this large natural variability, the potential long-term trends are evident in some regions, even effecting the local rainfall.
In this section, we will explore the relationship between rainfall and bushfires, is it positive or negative?

As a first step, we observe annual rainfall level that retrieved from the rainfall dataset. The interval of observation is ranging from 2000 until 2020.

Figure 4.8: Annual total rainfall trends from 2001 to 2019

Annual Rainfall and Bushfires

Figure 4.9: Annual Rainfall and Bushfires

Figure 4.9: Annual Rainfall and Bushfires

Similarly with our previous job when comparing the trend of average temperature to bushfires, we will rescale down the annual average fire spots using log 10 in order to put its trend side by side with average rainfall trend in the same plot.

As shown in figure 4.9, we can see that, in general, annual rainfall has a decreasing trend during the observed period (2000 - 2020), while bushfires has a slightly rising trend. Hence, we can assume that rainfall and bushfires are negatively associated. This contrast pattern is quite obvious particularly in 2011 where average fire spots soar as the average rainfall level drops from 4.9 mm in 2010 to 4.15 mm in 2011. After 2015, bushfires is slightly increasing until 2019 as rainfall level continues to drop to its lowest level (2.16 mm) in the same year.

To complete this analysis, we will check the exact coefficient number of correlation between variable rainfall and bushfires in the following section.

4.3.3 The correlation between climatic conditions and bushfires

From Section 4.3.1 and 4.3.2, we know that temperature, rainfall, and fire are related, but how closely are they related? In this part, we mainly explore the correlation coefficient between them.

The correlation between rainfall, temperature and bushfires

Figure 4.10: The correlation between rainfall, temperature and bushfires

From Figure 4.10, it seems that the rainfall and bushfires have negative correlation, suggesting that the more rainfall and the less bushfires. On the contrary, the higher temperature and the more bushfires, although the correlation between these two variables is slightly small, while the relationship between them still can be evaluated. Moreover, the rainfall will be declined when the temperature would get higher. Temperatures will impact the rate of evaporation, with higher temperatures leading to faster soil moisture loss(Hausfather 2018).

In conclusion, the main climatic conditions of bushfire is hot and dry(Oldenborgh et al. 2020). The graph of annual rainfall and temperature show that the hottest and driest year is 2019, which has the most massive bushfires in Australia as well. Therefore, as the combination of arid and severe hot conditions adds up to more powerful fires, indicating that declines in rainfall and increases in temperature have likely been a primary driver of increases in wildfire area burned.

5 Conclusions

This research briefly analyzes when and where bushfires are more likely to occur and the climate conditions under which they occur. And we find that the bushfires are prone to the month from August and November every year, and the Northern Territory and Western Australia are the most prone to fires. Moreover, high temperature and drought are critical climatic conditions for the occurrence of bushfires. For future research, it will be interesting to explore how to prevent bushfires in terms of climate conditions and to analyze the effects of climate change in bushfires.

Acknowlegments

The authors would like to thank all the contributors to the following R package: Wickham et al. (2019), Wickham (2016), Wickham, Hester, and Francois (2018), Cheng, Karambelkar, and Xie (2019), Ryan and Ulrich (2020), Müller (2017), R Core Team (2020), Arnold (2019), Wickham et al. (2020), Vanderkam et al. (2018), Grolemund and Wickham (2011), Schloerke et al. (2020), Rudis (2020).

References

Arnold, Jeffrey B. 2019. Ggthemes: Extra Themes, Scales and Geoms for ’Ggplot2’. https://CRAN.R-project.org/package=ggthemes.

Australia, Geoscience. 2020. “Bushfire.” https://www.ga.gov.au/scientific-topics/community-safety/bushfire.

Cheng, Joe, Bhaskar Karambelkar, and Yihui Xie. 2019. Leaflet: Create Interactive Web Maps with the Javascript ’Leaflet’ Library. https://CRAN.R-project.org/package=leaflet.

Grolemund, Garrett, and Hadley Wickham. 2011. “Dates and Times Made Easy with lubridate.” Journal of Statistical Software 40 (3): 1–25. http://www.jstatsoft.org/v40/i03/.

Hausfather, Zeke. 2018. “Explainer: What Climate Models Tell Us About Future Rainfall.” Carbon Brief.

Müller, Kirill. 2017. Here: A Simpler Way to Find Your Files. https://CRAN.R-project.org/package=here.

NASA. 2020. “NASA Fires of Australia.”

Oldenborgh, Geert Jan van, Folmer Krikken, Sophie Lewis, Nicholas J Leach, Flavio Lehner, Kate R Saunders, Michiel van Weele, et al. 2020. “Attribution of the Australian Bushfire Risk to Anthropogenic Climate Change.” Natural Hazards and Earth System Sciences Discussions, 1–46.

R Core Team. 2020. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.

Rudis, Bob. 2020. Hrbrthemes: Additional Themes, Theme Components and Utilities for ’Ggplot2’. https://CRAN.R-project.org/package=hrbrthemes.

Ryan, Jeffrey A., and Joshua M. Ulrich. 2020. Xts: EXtensible Time Series. https://CRAN.R-project.org/package=xts.

Schloerke, Barret, Di Cook, Joseph Larmarange, Francois Briatte, Moritz Marbach, Edwin Thoen, Amos Elberg, and Jason Crowley. 2020. GGally: Extension to ’Ggplot2’. https://CRAN.R-project.org/package=GGally.

Sullivan, Andrew L, W Lachie McCaw, Miguel G Cruz, Stuart Matthews, and Peter F Ellis. 2012. “Fuel, Fire Weather and Fire Behaviour in Australian Ecosystems.” Flammable Australia: Fire Regimes, Biodiversity and Ecosystems in a Changing World, 51–77.

Vanderkam, Dan, JJ Allaire, Jonathan Owen, Daniel Gromer, and Benoit Thieurmel. 2018. Dygraphs: Interface to ’Dygraphs’ Interactive Time Series Charting Library. https://CRAN.R-project.org/package=dygraphs.

Wickham, Hadley. 2016. Ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York. https://ggplot2.tidyverse.org.

Wickham, Hadley, Mara Averick, Jennifer Bryan, Winston Chang, Lucy D’Agostino McGowan, Romain François, Garrett Grolemund, et al. 2019. “Welcome to the tidyverse.” Journal of Open Source Software 4 (43): 1686. https://doi.org/10.21105/joss.01686.

Wickham, Hadley, Romain François, Lionel Henry, and Kirill Müller. 2020. Dplyr: A Grammar of Data Manipulation. https://CRAN.R-project.org/package=dplyr.

Wickham, Hadley, Jim Hester, and Romain Francois. 2018. Readr: Read Rectangular Text Data. https://CRAN.R-project.org/package=readr.